Mirix is a multi-agent personal assistant designed to capture and structure on-screen activities into a local, adaptable knowledge base. It targets end-users and developers seeking to build personalized AI memory systems, offering intelligent conversation and advanced search capabilities over captured digital experiences.

How It Works

Mirix employs a multi-agent architecture with six specialized memory components (Core, Episodic, Semantic, Procedural, Resource, Knowledge Vault). Dedicated agents manage these components, processing screen activity and multi-modal inputs (text, images, voice, screen captures) to consolidate information into structured memories. This approach allows for a rich, adaptive knowledge base that can be queried via advanced search, including PostgreSQL's BM25 and vector similarity.

Quick Start & Requirements

• Installation: Clone the repository (git clone git@github.com:Mirix-AI/MIRIX.git), create and activate a virtual environment, then run pip install -r requirements.txt. Alternatively, use the Python SDK: pip install mirix.
• Prerequisites: Python, PostgreSQL (for advanced search), and an API key for the default memory agent (Google Gemini 2.0 Flash).
• Resources: Requires local storage for data and potentially significant compute for processing screen activity and AI models.
• Links: Website, Documentation, Paper

Highlighted Details

• Multi-agent memory system with six specialized components.
• Continuous screen activity tracking and consolidation into structured memories.
• Privacy-first design with all long-term data stored locally.
• Advanced search combining PostgreSQL BM25 and vector similarity.

Maintenance & Community

• Active development with a public GitHub repository.
• Community channels available via Discord Community and WeChat.

Licensing & Compatibility

• Released under the Apache License 2.0.
• Permissive license suitable for commercial use and integration into closed-source applications.

Limitations & Caveats

The project acknowledges being built upon Letta's open-sourced framework, suggesting potential dependencies or architectural similarities. Specific hardware requirements for optimal screen tracking performance are not detailed.